PERFORCE change 100719 for review

[John Baldwin]

On Thursday 06 July 2006 07:05, Roman Divacky wrote:
> http://perforce.freebsd.org/chv.cgi?CH=100719
> 
> Change 100719 by rdivacky at rdivacky_witten on 2006/07/06 11:04:26
> 
> 	TID handling
> 	o	introduction of P_LINUX proc flag
> 	o	introduction of linux hooks into exit1() and userret()
> 	o	TID handling in i386 version of linuxolator
> 	
> 	this has not been tested much. More testing will come once the futexes are finished.
> 
> Affected files ...
> 
> .. //depot/projects/soc2006/rdivacky_linuxolator/i386/linux/linux_dummy.c#5 edit
> .. //depot/projects/soc2006/rdivacky_linuxolator/i386/linux/linux_machdep.c#7 edit
> .. //depot/projects/soc2006/rdivacky_linuxolator/i386/linux/linux_proto.h#8 edit
> .. //depot/projects/soc2006/rdivacky_linuxolator/i386/linux/linux_syscall.h#8 edit
> .. //depot/projects/soc2006/rdivacky_linuxolator/i386/linux/linux_sysent.c#8 edit
> .. //depot/projects/soc2006/rdivacky_linuxolator/i386/linux/linux_sysvec.c#5 edit
> .. //depot/projects/soc2006/rdivacky_linuxolator/i386/linux/syscalls.master#8 edit
> .. //depot/projects/soc2006/rdivacky_linuxolator/kern/kern_exit.c#3 edit
> .. //depot/projects/soc2006/rdivacky_linuxolator/kern/subr_trap.c#2 edit
> .. //depot/projects/soc2006/rdivacky_linuxolator/sys/proc.h#2 edit
> 
> Differences ...
> 
> ==== //depot/projects/soc2006/rdivacky_linuxolator/kern/kern_exit.c#3 (text+ko) ====
> 
> @@ -87,6 +87,7 @@
>  
>  /* Hook for NFS teardown procedure. */
>  void (*nlminfo_release_p)(struct proc *p);
> +int (*linux_proc_exit_p)(struct thread *) = NULL;	/* necessary for linuxolator */
>  
>  /*
>   * exit --
> @@ -236,6 +237,9 @@
>  	 */
>  	EVENTHANDLER_INVOKE(process_exit, p);
>  
> +	/* we have to call linux exit hook */
> +	if (p->p_flag & P_LINUX && linux_proc_exit_p != NULL)
> +	   	(linux_proc_exit_p)(td);
>  	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
>  		M_ZOMBIE, M_WAITOK);
>  	/*
> 

Please use a 'process_exit' eventhandler instead.  You can test P_LINUX in
your handler and return right away for non-Linux procs.

> ==== //depot/projects/soc2006/rdivacky_linuxolator/kern/subr_trap.c#2 (text+ko) ====
> 
> @@ -67,6 +67,8 @@
>  #include <machine/cpu.h>
>  #include <machine/pcb.h>
>  
> +int (*linux_userret_p)(struct thread *) = NULL;	/* for linuxolator */
> +
>  /*
>   * Define the code needed before returning to user mode, for
>   * trap and syscall.
> @@ -128,6 +130,10 @@
>  		addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio);
>  	}
>  
> +	/* linux userret */
> +	if (p->p_flag & P_LINUX && linux_userret_p != NULL)
> +	   (linux_userret_p)(td);
> +
>  	/*
>  	 * Let the scheduler adjust our priority etc.
>  	 */
> 

If it is needed we can add a new 'thread_userret' eventhandler rather than
putting Linux-specific hooks into the code.

> ==== //depot/projects/soc2006/rdivacky_linuxolator/sys/proc.h#2 (text+ko) ====
> 
> @@ -655,6 +655,7 @@
>  #define	P_HWPMC		0x800000 /* Process is using HWPMCs */
>  
>  #define	P_JAILED	0x1000000 /* Process is in jail. */
> +#define P_LINUX		0x2000000 /* linux binary */
>  #define	P_INEXEC	0x4000000 /* Process is in execve(). */
>  #define	P_STATCHILD	0x8000000 /* Child process stopped or exited. */

If you do the two above, then you might be able to come up with another test
(compare p_sysvec?) instead of needing this flag since the only tests of it
would be in the emulator code itself.

-- 
John Baldwin